Modern marine ecology, particularly the study of benthic ecosystems, is increasingly driven by the acquisition of large volumes of imagery. From towed camera systems, remotely operated vehicles and autonomous underwater vehicles, underwater imaging systems now capture vast quantities of visual data. However, the interpretation of this imagery is slow, costly, and often inconsistent, creating a fundamental bottleneck that restricts the scale and speed of ecological insight.
Recent advances in computer vision methods can help to clear this bottleneck, assisting with tasks like taxonomic identification, abundance estimates, and habitat mapping. Computer vision methods have the potential to transform how benthic ecosystems are studied, enabling ecologists to work faster, more consistently, and at far greater scales. Despite this promise, many benthic ecologists do not understand the methods underpinning computer vision, lowering confidence in their application. Additionally, marine imagery presents distinct complexities for computer vision, namely the presence of turbidity, variable lighting, and cluttered scenes which cause “off-the-shelf" models to perform poorly. The lack of core computer vision skills and the unique challenges posed by underwater imagery has slowed the marine science community’s ability to adopt, adapt and harness computer vision. This workshop will address these challenges directly. It will equip benthic ecologists, primarily early-career researchers with little programming experience and no prior machine learning background, with the foundational knowledge, skills, and confidence required to integrate computer vision into their research.
The workshop will run over two days, hosted in-person at the British Antarctic Survey in Cambridge, 10–11th March 2026, from 10:00 - 17:00 on the 10th and 09:00 - 16:00 on the 11th. Teaching will be delivered through two complementary elements: accessible lectures introducing key concepts, and guided hands-on practicals using a real benthic dataset.
Day 1 – Foundations
- Morning lectures: Introduction to computer vision; challenges in marine imagery; effective data labelling.
- Afternoon practical: Building a benthic image dataset.
- Evening: Networking reception and dinner.
Day 2 – Applications
- Morning lectures: Model training, evaluation, and working with regulatory/industry end-users.
- Afternoon practical: Training and evaluating models using the dataset from Day 1.
By the end of this workshop participants will be able to:
- Describe the core concepts underlying modern computer vision (e.g. classification, object detection, segmentation) and how they apply to benthic imagery.
- Identify when and how computer vision is an appropriate tool for a specific research question.
- Recognise the distinct challenges of working with marine data (e.g. turbidity, colour distortion) and how they affect computer vision systems.
- Annotate benthic imagery effectively and understand how labelling types impact model performance.
- Outline best-practice workflows for computer vision, from dataset preparation and annotation to training and evaluation.
- Train and assess a computer vision model using open-source, reproducible code.
- Have the knowledge and resource to apply computer vision to their own marine image analysis problem.